{"id":620640,"date":"2023-03-22T09:49:13","date_gmt":"2023-03-22T14:49:13","guid":{"rendered":"https:\/\/news.sellorbuyhomefast.com\/index.php\/2023\/03\/22\/can-ai-generated-text-be-reliably-detected\/"},"modified":"2023-03-22T09:49:13","modified_gmt":"2023-03-22T14:49:13","slug":"can-ai-generated-text-be-reliably-detected","status":"publish","type":"post","link":"https:\/\/newsycanuse.com\/index.php\/2023\/03\/22\/can-ai-generated-text-be-reliably-detected\/","title":{"rendered":"Can AI-Generated Text Be Reliably Detected?"},"content":{"rendered":"<div id=\"content-inner\">\n<p>  [Submitted on 17 Mar 2023]<\/p>\n<p><a aria-describedby=\"download-button-info\" href=\"http:\/\/arxiv.org\/pdf\/2303.11156\">Download PDF<\/a><\/p>\n<blockquote><p>\n      <span>Abstract:<\/span>  The rapid progress of Large Language Models (LLMs) has made them capable of<br \/>\nperforming astonishingly well on various tasks including document completion<br \/>\nand question answering. The unregulated use of these models, however, can<br \/>\npotentially lead to malicious consequences such as plagiarism, generating fake<br \/>\nnews, spamming, etc. Therefore, reliable detection of AI-generated text can be<br \/>\ncritical to ensure the responsible use of LLMs. Recent works attempt to tackle<br \/>\nthis problem either using certain model signatures present in the generated<br \/>\ntext outputs or by applying watermarking techniques that imprint specific<br \/>\npatterns onto them. In this paper, both empirically and theoretically, we show<br \/>\nthat these detectors are not reliable in practical scenarios. Empirically, we<br \/>\nshow that paraphrasing attacks, where a light paraphraser is applied on top of<br \/>\nthe generative text model, can break a whole range of detectors, including the<br \/>\nones using the watermarking schemes as well as neural network-based detectors<br \/>\nand zero-shot classifiers. We then provide a theoretical impossibility result<br \/>\nindicating that for a sufficiently good language model, even the best-possible<br \/>\ndetector can only perform marginally better than a random classifier. Finally,<br \/>\nwe show that even LLMs protected by watermarking schemes can be vulnerable<br \/>\nagainst spoofing attacks where adversarial humans can infer hidden watermarking<br \/>\nsignatures and add them to their generated text to be detected as text<br \/>\ngenerated by the LLMs, potentially causing reputational damages to their<br \/>\ndevelopers. We believe these results can open an honest conversation in the<br \/>\ncommunity regarding the ethical and reliable use of AI-generated text.<\/p>\n<\/blockquote><\/div>\n<div>\n<h2>Submission history<\/h2>\n<p> From: Aounon Kumar [<a href=\"http:\/\/arxiv.org\/show-email\/f72655cd\/2303.11156\">view email<\/a>]<br \/>\n      <br \/><strong>[v1]<\/strong><br \/>\nFri, 17 Mar 2023 17:53:19 UTC (926 KB)<\/p>\n<\/div>\n<p><a href=\"https:\/\/arxiv.org\/abs\/2303.11156\" class=\"button purchase\" rel=\"nofollow noopener\" target=\"_blank\">Read More<\/a><br \/>\n Raleigh Menjivar<\/p>\n","protected":false},"excerpt":{"rendered":"<p>[Submitted on 17 Mar 2023] Download PDF Abstract: The rapid progress of Large Language Models (LLMs) has made them capable of performing astonishingly well on various tasks including document completion and question answering. The unregulated use of these models, however, can potentially lead to malicious consequences such as plagiarism, generating fake news, spamming, etc. Therefore<\/p>\n","protected":false},"author":1,"featured_media":620641,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[42579,26041,46],"tags":[],"class_list":{"0":"post-620640","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-ai-generated","8":"category-reliably","9":"category-technology"},"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts\/620640","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/comments?post=620640"}],"version-history":[{"count":0,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/posts\/620640\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/media\/620641"}],"wp:attachment":[{"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/media?parent=620640"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/categories?post=620640"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/newsycanuse.com\/index.php\/wp-json\/wp\/v2\/tags?post=620640"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}